Lamy publication on the protection of creations generated by artificial intelligence.
The American position, which must however be nuanced since the dismissal of the Director of the U.S. Copyright Office by the President of the United States on 10 May, is of particular importance for French lawyers, as it comes in a context where the European Union is simultaneously developing its own regulatory framework for AI.
The central question that is emerging is that of determining the conditions under which creations involving AI can benefit from copyright or copyright protection. This problem requires examining both the technical aspects of the creative process and the conceptual foundations of copyright.
I) The criteria for the protection of content involving generative AI
The U.S. Copyright Office states in its report that copyright principles are flexible enough to apply to generative AIs and to protect creations that involve this technology in whole or in part.
Several use cases of generative AI are then detailed, as well as the criteria that allow copyright protection of generated content.
A) The contents not protectable: those generated solely on the basis of a text prompt
The Agency concludes that prompts (written instructions to generative AI systems to describe the desired result) do not provide sufficient human control of the result generated by AI. As a result, the U.S. Copyright Office declares that content generated by AI based solely on text prompts is not eligible for copyright protection.
This position is based mainly on the technical argument of the unpredictability inherent in the AI generation process: a single prompt is subject to the interpretation of the AI model used and is likely to generate a large number of different results, over which the human creator has no control.
This position of the U.S. Copyright Office therefore implies that in the current state of functioning of generative AIs, prompts alone are not sufficient to demonstrate human creative effort. Such a conclusion, motivated by purely technical arguments, could be followed in other countries and in particular in France. This implies that it would be well advised to keep material evidence that is more diverse than just prompt ones in order to be in a position to prove the existence of human creative effort in the event of a dispute over the existence of copyright.
B) Conditionally protected content: content retouched by post-generation AI
According to the U.S. Copyright Office, AI tools that make it possible to modify outputs can demonstrate human control of the creative process and therefore justify copyright protection of the output. This copyright protection will be granted on a case-by-case basis, after examining all retouching steps.
In particular, the Office specified that the “vary region” and “remix mode” tools available for the generative AI model marketed by Midjourney can demonstrate human control of the creative process and therefore justify copyright protection of the output. Services for the “upscaling” of content generated by AI could also make it possible, in some cases, to demonstrate human creative effort, according to the Office.
C) Protectable content: content generated using reference elements
The Office distinguishes between text prompts and “expressive prompts,” such as a reference image injected into the generative AI tool in order to generate new content based on this. In the event that the expressive prompt testifies to a human creative effort and that this effort is perceptible in the generated result, the visible portion of the reference content is likely to be protected by copyright.
The scope of protection could therefore be analogous to that of a derivative work, so that any copyright in an AI-generated result based on reference content would be limited to the part where the expression of human creative effort is visible.
II) Practical and strategic implications
Creators and businesses need to adapt their practices to secure the protection of AI-generated content. The analysis of American positions reveals concrete strategies to adopt, while anticipating technological developments that could redefine protection criteria. These practical recommendations are based on a technical understanding of current AI tools and emerging case law requirements.
A) The good ones practices for creators to adopt
- Documenting The creative process
Preserving physical evidence of human control over the creative process is becoming crucial. This includes documenting creative steps, artistic choices, and changes to AI-generated content. The “A Single Piece of American Cheese” case of January 30, 2025 illustrates this approach: the U.S. Copyright Office granted copyright protection to the digital work, because the artist had demonstrated his creative intervention by providing a timelapse of the creation and a detailed explanation of his role in the process.
- Don't just rely on the quick ones to prove human creative effort
According to the U.S. Copyright Office, prompts are not enough, in the current state of technology, to prove human creative effort because a prompt is subject to machine interpretation and can lead to a multitude of different results, over which humans have no control. This conclusion by the U.S. Copyright Office, based on purely technical findings, could influence other countries, in particular France. It is therefore strongly recommended not to limit yourself to keeping prompts alone in order to document the creative process. It is advisable to keep more varied elements such as preparatory sketches, successive modifications or artistic choices in order to be able to defend one's copyright in the event of a dispute.
- Use proprietary reference elements
Injecting reference content whose author owns the rights as instructions for the AI system is an effective strategy for securing the protection of AI-generated content, provided that these elements remain noticeable in the result. This approach makes it possible to create a “protected base” on which AI-assisted creation is based. For example, a graphic designer can start from one of their existing drawings in order to generate content by AI based on this, then make personal adjustments to it. The presence of identifiable human creative elements in the final content greatly enhances the chances of obtaining copyright protection — and copyright protection for France.
- Limit the transfer of rights to 100% AI-generated content
It is important to avoid transferring rights to content whose protectability remains uncertain, especially for content generated by simple prompting. This caution is necessary because giving up non-existent or questionable rights can create major legal complications and weaken the creator's contractual position. For France, it is preferable to await legal clarification on this point and to limit transfers to works where creative human intervention is clearly documented and substantial. This approach protects creators against possible contractual liability claims while maintaining their rights to creations whose legal value could be recognized later.
B) Prospects for the evolution of the legal framework
The U.S. Copyright Office recognizes that its findings depend on the current state of technology and that new advancements could change the conditions under which AI-generated content is protected by copyright. By objectively evaluating the current capabilities of AI tools, which currently offer limited control to the user over the final result, the Office adopts a factual approach that can be transposed into other legal systems, including France.
Contrary to legal principles that vary from country to country, the assessment of the technological capabilities of an AI remains the same regardless of the jurisdiction. The French authorities could therefore logically reach the same conclusions as their American counterparts, based on a shared technical observation. This position on a technological argument highlights the importance of continuous monitoring: future generations of generative AI tools can substantially change the balance between human control and artificial autonomy and thus their eligibility for protection.
Link to the article to be consulted on the Lamy Liaisons website